Pain is a complex sensory and emotional experience, often difficult to assess objectively. In recent years, artificial intelligence (AI) has shown great potential in improving the accuracy and efficiency of pain assessment. This study aims to conduct a systematic review of AI-based pain detection methods developed in the period 2020 to 2024. Using the PRISMA 2020 approach, a literature search was conducted in three major databases: PubMed, Scopus, and Google Scholar, with keywords related to pain detection and perception. Of the 1,685 articles found, 44 studies were selected through a rigorous selection process. The analysis of five showed the main approaches in pain detection: Neuroimaging & Neurological, Physiological & Biometric, Visual-Only (facial recognition), Audio/Speech-based, and Behavioral/Observational. Neuroimaging-based approaches such as EEG and fMRI were the most dominant, followed by the use of biometric sensors and facial recognition technology. However, significant challenges remain, including the limitations of global data standards, difficulties in model generalization, and ethical and privacy issues. This study highlights that the integration of non-invasive sensors with deep learning models and personalized approaches can improve the effectiveness of automated pain detection systems.